def block_generator(blocks, depth, limits, sparsity, jitter, empty):
output_blocks = []
for b in range(blocks):
output_blocks.append([])
b_limits = (int(norm(limits[0], jitter)), int(norm(limits[1], jitter)))
current = b_limits[0]
output_blocks[b].append(
[b_limits[0]] +
[10 if rand() < empty else randint(11, 127) for i in range(depth)])
while forb(current, b_limits[1],
40) >= 0: # While we have not passed the upper limit
current = (current + (sparsity + 1)) % 1024
output_blocks[b].append([current] + [
10 if rand() < empty else randint(11, 127)
for i in range(depth)
])
if b % 2 == 1:
output_blocks[b] = output_blocks[b][::-1]
return output_blocks
def init_weights(self, act):
ret = []
for n, layer in enumerate(act[:-1]):
layer = layer[0]
ret.append([])
next_layer = act[n + 1][0]
for node_1 in range(len(layer)):
ret[-1].append([])
for node_2 in range(len(next_layer)):
ret[-1][-1].append(norm(0, 1))
return ret
def genEnsemble(self, phase = 'allyears'): from random import normalvariate as norm from numpy import zeros sd = self.hindcast_error[phase].std() mn = self.hindcast_error[phase].mean() n = len(self.hindcast[phase]) ensemble = zeros((1000,n)) for i in range(n): for j in range(1000): y = norm(mn, sd) ensemble[j,i] = self.hindcast[phase][i] + y self.ensemble[phase] = ensemble return
def brownian_bridge(ti, tf, bi, bf, n):
"""
simulation d'un pont brownien sur [ti,tf],
avec les valeurs extrèmes bi et bf
et n points par unité de temps
sortie :
- T : positions temporelles des échantillons
- B : valeurs des échantillons
"""
n = int(n*(tf-ti)) # nombre de points
T = linspace(ti,tf,n) # points d'échantillonnage
pas = (tf-ti)/(n-1.) # pas d'échantillonnage
B = zeros(n) # initialisation du brownien
B[0] = bi # valeur initiale
B[n-1] = bf # valeur finale
t1 = ti
for k in range(1,n-1): # construction du pont en ti+k*pas
m = moy(t1,tf,B[k-1],bf,t1+pas) # sur les intervalle [ti+(k-1)*pas,tf]
v = var(t1,tf,B[k-1],bf,t1+pas) # avec les valeurs limites B[k-1],et bf
B[k] = m+sqrt(v)*norm(0,1)
t1 += pas
return T, B
def int_range(mean, std_dev=None, max_std_dev=2):
"Return an random integer normally distributed around mean, with the given std dev."
if std_dev is None:
std_dev = mean / 4.0
mean += 0.5
return int(min(mean+std_dev*max_std_dev, max(norm(mean, std_dev), mean-std_dev*max_std_dev)))
